Evaluation of Virtual Screening Performance of Support Vector Machines Trained by Sparsely Distributed Active Compounds

Abstract

Virtual screening performance of support vector machines (SVM) depends on the diversity of training active and inactive compounds. While diverse inactive compounds can be routinely generated, the number and diversity of known actives are typically low. We evaluated the performance of SVM trained by sparsely distributed actives in six MDDR biological target classes composed of a high number of known actives (983-1645) of high, intermediate, and low structural diversity (muscarinic M1 receptor agonists, NMDA receptor antagonists, thrombin inhibitors, HIV protease inhibitors, cephalosporins, and renin inhibitors). SVM trained by regularly sparse data sets of 100 actives show improved yields at substantially reduced false-hit rates compared to those of published studies and those of Tanimoto-based similarity searching method based on the same data sets and molecular descriptors. SVM trained by very sparse data sets of 40 actives (2.4%-4.1% of the known actives) predicted 17.5-39.5%, 23.0-48.1%, and 70.2-92.4% of the remaining 943-1605 actives in the high, intermediate, and low diversity classes, respectively, 13.8-68.7% of which are outside the training compound families. SVM predicted 99.97% and 97.1% of the 9.997 M PUBCHEM and 167K remaining MDDR compounds as inactive and 2.6%-8.3% of the 19,495-38,483 MDDR compounds similar to the known actives as active. These suggest that SVM has substantial capability in identifying novel active compounds from sparse active data sets at low false-hit rates.

The influence of negative training set size on machine learning-based virtual screening

“…The vast majority of those methods require the preparation of a training set of compounds (supervised learning) that are used to develop a decision rule that can be then applied to sort a dataset of new molecules (the test set) among particular activity classes [1]. A number of studies have aimed to determine optimal learning parameters and examine their impact on classification effectiveness [2-5]. Interestingly, no extensive research that considers the influence of the ratio of active to inactive training examples on the efficiency of new active compounds recognition has been performed.…”

Consensus model for identification of novel PI3K inhibitors in large chemical library

“…2, resulting in a model that cannot identify an active compound that has similar structure to the putative negative compounds. The extent of this risk is unknown but the results of this work and two other studies [23,27] have shown that such unwanted effect is expected to be relatively small and it was still possible for a substantial proportion of positive compounds to be classified correctly despite their membership in negative families. Nonetheless, the search for known PI3K inhibitors in this work was carried out to be as extensive as possible to minimize this risk.…”

“…Thus, this study has adopted the approach by Han et al [19] to generate putative inactive compounds to augment the negative training set. This method can generate putative negatives without requiring the knowledge of actual inactive compounds and studies had shown that classification models derived from these putative negatives can perform reasonably well in virtual screening [23,27]. Nonetheless, the effects of using a large number of putative negatives was examined to ensure that the change is not unacceptably detrimental to the identification of potential inhibitors.…”

“…Voting was chosen for the consensus method in this study and three base classifiers, k-nearest neighbor (kNN), aggregating one-dependence estimators (AODE) which is a variant of naive Bayes, and support vector machine (SVM) which has gained popularity in recent years, were used. SVM is useful for the classification of systems where there is limited knowledge on the mechanism or specific association between the activities and molecular properties [22] because it classifies compounds on the basis of the discriminative properties between active and inactive compounds rather than structural similarity to active compounds unlike most of the non-machine learning methods [23]. Hence, it is expected that a consensus model which considers the prediction results from the three base classifiers will be useful for the virtual screening of potential PI3K inhibitors from large chemical libraries.…”

In-Silico Approaches to Multi-target Drug Discovery

“…In-silico methods have been widely explored for facilitating lead discovery against individual targets (33,34). In particular, molecular docking (35), pharmacophore (36), structure-activity relationship (SAR) and quantitative structure activity relationship (QSAR) (37), machine learning (38), and combination methods (39) have been extensively used for searching and designing active compounds against individual targets.…”